structural_simplify on DiscreteSystem says "this should never happen"

[pepijndevos]

I can't really find much documentation on DiscreteSystem, except the example inside the docstring.

It seems that every LHS variable is treated as t+1, while I also want to have algebraic equations. I figured if I'd structural simplify the system it might simplify the algebraic equations leaving only the difference equations. But instead it just throws an exception.

using ModelingToolkit

@parameters t σ ρ β
@variables x(t) y(t) z(t) next_x(t) next_y(t) next_z(t)

eqs = [next_x ~ σ*(y-x),
       next_y ~ x*(ρ-z)-y,
       next_z ~ x*y - β*z]

de = DiscreteSystem(eqs,t,[x,y,z],[σ,ρ,β])
structural_simplify(de)

ERROR: LoadError: This should never happen. Trying to set DataType with NamedTuple{(:structure,), Tuple{SystemStructure}}.
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:33
  [2] #s58#28
    @ ~/.julia/packages/ModelingToolkit/yGWxM/src/systems/abstractsystem.jl:200 [inlined]
  [3] var"#s58#28"(::Any, obj::Any, patch::Any)
    @ ModelingToolkit ./none:0
  [4] (::Core.GeneratedFunctionStub)(::Any, ::Vararg{Any, N} where N)
    @ Core ./boot.jl:571
  [5] set
    @ ~/.julia/packages/Setfield/NshXm/src/lens.jl:110 [inlined]
  [6] macro expansion
    @ ~/.julia/packages/Setfield/NshXm/src/sugar.jl:192 [inlined]
  [7] initialize_system_structure(sys::DiscreteSystem)
    @ ModelingToolkit.SystemStructures ~/.julia/packages/ModelingToolkit/yGWxM/src/systems/systemstructure.jl:198
  [8] alias_elimination(sys::DiscreteSystem)
    @ ModelingToolkit ~/.julia/packages/ModelingToolkit/yGWxM/src/systems/alias_elimination.jl:6
  [9] structural_simplify(sys::DiscreteSystem)
    @ ModelingToolkit ~/.julia/packages/ModelingToolkit/yGWxM/src/systems/abstractsystem.jl:767
 [10] top-level scope
    @ ~/code/juliacomp/symbolichdl.jl/src/discrete.jl:18
in expression starting at /home/pepijn/code/juliacomp/symbolichdl.jl/src/discrete.jl:18

[ChrisRackauckas]

you didn't define a differencing operator, so these equations don't make much sense.

[ChrisRackauckas]

https://github.com/SciML/ModelingToolkit.jl/blob/master/test/discretesystem.jl

[pepijndevos]

I copied this from the docstring for DiscreteSystem so then that needs updating.

[pepijndevos]

Well, adding that doesn't make much difference. Simplificaiton on discrete systems is just broken

using ModelingToolkit

@parameters t σ ρ β
@variables x(t) y(t) z(t) next_x(t) next_y(t) next_z(t)

D = Difference(t; dt=1)

eqs = [
    D(x) ~ σ*(y-x),
    D(y) ~ x*(ρ-z)-y,
    D(z) ~ x*y - β*z
]

de = DiscreteSystem(eqs,t,[x,y,z],[σ,ρ,β])
structural_simplify(de)

Same error.

It also doesn't really do algebraic equations.

eqs = [
    D(x) ~ x + 1,
    y ~ x,
    z ~ y
]

I'd expect x, y, and x to be equal here, but instead it behaves as if it all 3 were difference equations.

 [1, 1, 1]
 [2, 1, 1]
 [3, 2, 1]
 [4, 3, 2]
 [5, 4, 3]
 [6, 5, 4]
 [7, 6, 5]
 [8, 7, 6]
 [9, 8, 7]
 [10, 9, 8]
 [11, 10, 9]

[baggepinnen]

The error reported in the OP

ERROR: LoadError: This should never happen. Trying to set DataType with NamedTuple{(:structure,), Tuple{SystemStructure}}.
Stacktrace:

is due to DiscreteSystems not having the field structure which is assumed by structural_simplify.

@YingboMa would it be appropriate to add the field structure to DiscreteSystem, or these systems need completely different treatment?

[ChrisRackauckas]

I think all of the dependency analyses are fine and carry over to the DiscreteSystem case, so adding structure so that structural_simplify works should be what's needed.

[ChrisRackauckas]

DiscreteSystem was removed. A new discrete system is being developed.